Open source by default

The 10 Clawproof checks are open source (MIT) and packaged as a Claude skill. Clone the repo, audit your agent, run the skill inside Claude Code, or propose a new check from your own production failures.

Run the production gate

# Clone the repo
git clone https://github.com/lexbeam-software/clawproof-checks.git
cd clawproof-checks

# Start from the example declaration and add your evidence references
cp examples/gate-review.json clawproof-gate.json
./bin/clawproof gate clawproof-gate.json --format markdown --target my-agent

# Exit codes: PASS=0, REVIEW=2, BLOCK=3, INVALID=4

The gate is deterministic and makes no network calls. A critical failed control blocks release even when the coverage score is high. A yes answer without an evidence reference requires review.

Use the audit skill

# From the cloned repository, copy the skill into your project
mkdir -p /path/to/your-project/.claude/skills
cp -r skills/clawproof-audit /path/to/your-project/.claude/skills/

# Then ask Claude Code to run the clawproof-audit skill.

Use it programmatically

Every check is bundled into a single JSON file (checks.json) for programmatic use outside Claude. Read the schema at checks/SCHEMA.md.

For the complete browser-readable version, use the AI agent production readiness checklist.

Roadmap

• ✓ 10 checks as YAML (v1)
• ✓ Claude skill + audit-report template
• ✓ Bundled checks.json for programmatic use
• ✓ Deterministic evidence gate with CI exit codes
• □ MCP server exposinglist_checks andrun_check tools (planned)
• □ Language-specific playbooks (PRs welcome)
• □ German and French translations (PRs welcome)